2 * Linux Socket Filter - Kernel level socket filtering
5 * Jay Schulist <jschlst@samba.org>
7 * Based on the design of:
8 * - The Berkeley Packet Filter
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or (at your option) any later version.
15 * Andi Kleen - Fix a few bad bugs and races.
16 * Kris Katterjohn - Added many additional checks in sk_chk_filter()
19 #include <linux/module.h>
20 #include <linux/types.h>
22 #include <linux/fcntl.h>
23 #include <linux/socket.h>
25 #include <linux/inet.h>
26 #include <linux/netdevice.h>
27 #include <linux/if_packet.h>
28 #include <linux/gfp.h>
30 #include <net/protocol.h>
31 #include <net/netlink.h>
32 #include <linux/skbuff.h>
34 #include <linux/errno.h>
35 #include <linux/timer.h>
36 #include <asm/uaccess.h>
37 #include <asm/unaligned.h>
38 #include <linux/filter.h>
39 #include <linux/reciprocal_div.h>
40 #include <linux/ratelimit.h>
41 #include <linux/seccomp.h>
43 /* No hurry in this branch
45 * Exported for the bpf jit load helper.
47 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, int k, unsigned int size)
52 ptr = skb_network_header(skb) + k - SKF_NET_OFF;
53 else if (k >= SKF_LL_OFF)
54 ptr = skb_mac_header(skb) + k - SKF_LL_OFF;
56 if (ptr >= skb->head && ptr + size <= skb_tail_pointer(skb))
61 static inline void *load_pointer(const struct sk_buff *skb, int k,
62 unsigned int size, void *buffer)
65 return skb_header_pointer(skb, k, size, buffer);
66 return bpf_internal_load_pointer_neg_helper(skb, k, size);
70 * sk_filter - run a packet through a socket filter
71 * @sk: sock associated with &sk_buff
72 * @skb: buffer to filter
74 * Run the filter code and then cut skb->data to correct size returned by
75 * sk_run_filter. If pkt_len is 0 we toss packet. If skb->len is smaller
76 * than pkt_len we keep whole skb->data. This is the socket level
77 * wrapper to sk_run_filter. It returns 0 if the packet should
78 * be accepted or -EPERM if the packet should be tossed.
81 int sk_filter(struct sock *sk, struct sk_buff *skb)
84 struct sk_filter *filter;
86 err = security_sock_rcv_skb(sk, skb);
91 filter = rcu_dereference(sk->sk_filter);
93 unsigned int pkt_len = SK_RUN_FILTER(filter, skb);
95 err = pkt_len ? pskb_trim(skb, pkt_len) : -EPERM;
101 EXPORT_SYMBOL(sk_filter);
104 * sk_run_filter - run a filter on a socket
105 * @skb: buffer to run the filter on
106 * @fentry: filter to apply
108 * Decode and apply filter instructions to the skb->data.
109 * Return length to keep, 0 for none. @skb is the data we are
110 * filtering, @filter is the array of filter instructions.
111 * Because all jumps are guaranteed to be before last instruction,
112 * and last instruction guaranteed to be a RET, we dont need to check
113 * flen. (We used to pass to this function the length of filter)
115 unsigned int sk_run_filter(const struct sk_buff *skb,
116 const struct sock_filter *fentry)
119 u32 A = 0; /* Accumulator */
120 u32 X = 0; /* Index Register */
121 u32 mem[BPF_MEMWORDS]; /* Scratch Memory Store */
126 * Process array of filter instructions.
129 #if defined(CONFIG_X86_32)
130 #define K (fentry->k)
132 const u32 K = fentry->k;
135 switch (fentry->code) {
136 case BPF_S_ALU_ADD_X:
139 case BPF_S_ALU_ADD_K:
142 case BPF_S_ALU_SUB_X:
145 case BPF_S_ALU_SUB_K:
148 case BPF_S_ALU_MUL_X:
151 case BPF_S_ALU_MUL_K:
154 case BPF_S_ALU_DIV_X:
159 case BPF_S_ALU_DIV_K:
160 A = reciprocal_divide(A, K);
162 case BPF_S_ALU_AND_X:
165 case BPF_S_ALU_AND_K:
174 case BPF_S_ALU_LSH_X:
177 case BPF_S_ALU_LSH_K:
180 case BPF_S_ALU_RSH_X:
183 case BPF_S_ALU_RSH_K:
192 case BPF_S_JMP_JGT_K:
193 fentry += (A > K) ? fentry->jt : fentry->jf;
195 case BPF_S_JMP_JGE_K:
196 fentry += (A >= K) ? fentry->jt : fentry->jf;
198 case BPF_S_JMP_JEQ_K:
199 fentry += (A == K) ? fentry->jt : fentry->jf;
201 case BPF_S_JMP_JSET_K:
202 fentry += (A & K) ? fentry->jt : fentry->jf;
204 case BPF_S_JMP_JGT_X:
205 fentry += (A > X) ? fentry->jt : fentry->jf;
207 case BPF_S_JMP_JGE_X:
208 fentry += (A >= X) ? fentry->jt : fentry->jf;
210 case BPF_S_JMP_JEQ_X:
211 fentry += (A == X) ? fentry->jt : fentry->jf;
213 case BPF_S_JMP_JSET_X:
214 fentry += (A & X) ? fentry->jt : fentry->jf;
219 ptr = load_pointer(skb, k, 4, &tmp);
221 A = get_unaligned_be32(ptr);
228 ptr = load_pointer(skb, k, 2, &tmp);
230 A = get_unaligned_be16(ptr);
237 ptr = load_pointer(skb, k, 1, &tmp);
246 case BPF_S_LDX_W_LEN:
258 case BPF_S_LDX_B_MSH:
259 ptr = load_pointer(skb, K, 1, &tmp);
261 X = (*(u8 *)ptr & 0xf) << 2;
293 case BPF_S_ANC_PROTOCOL:
294 A = ntohs(skb->protocol);
296 case BPF_S_ANC_PKTTYPE:
299 case BPF_S_ANC_IFINDEX:
302 A = skb->dev->ifindex;
307 case BPF_S_ANC_QUEUE:
308 A = skb->queue_mapping;
310 case BPF_S_ANC_HATYPE:
315 case BPF_S_ANC_RXHASH:
319 A = raw_smp_processor_id();
321 case BPF_S_ANC_ALU_XOR_X:
324 case BPF_S_ANC_NLATTR: {
327 if (skb_is_nonlinear(skb))
329 if (A > skb->len - sizeof(struct nlattr))
332 nla = nla_find((struct nlattr *)&skb->data[A],
335 A = (void *)nla - (void *)skb->data;
340 case BPF_S_ANC_NLATTR_NEST: {
343 if (skb_is_nonlinear(skb))
345 if (A > skb->len - sizeof(struct nlattr))
348 nla = (struct nlattr *)&skb->data[A];
349 if (nla->nla_len > A - skb->len)
352 nla = nla_find_nested(nla, X);
354 A = (void *)nla - (void *)skb->data;
359 #ifdef CONFIG_SECCOMP_FILTER
360 case BPF_S_ANC_SECCOMP_LD_W:
361 A = seccomp_bpf_load(fentry->k);
365 WARN_RATELIMIT(1, "Unknown code:%u jt:%u tf:%u k:%u\n",
366 fentry->code, fentry->jt,
367 fentry->jf, fentry->k);
374 EXPORT_SYMBOL(sk_run_filter);
378 * A BPF program is able to use 16 cells of memory to store intermediate
379 * values (check u32 mem[BPF_MEMWORDS] in sk_run_filter())
380 * As we dont want to clear mem[] array for each packet going through
381 * sk_run_filter(), we check that filter loaded by user never try to read
382 * a cell if not previously written, and we check all branches to be sure
383 * a malicious user doesn't try to abuse us.
385 static int check_load_and_stores(struct sock_filter *filter, int flen)
387 u16 *masks, memvalid = 0; /* one bit per cell, 16 cells */
390 BUILD_BUG_ON(BPF_MEMWORDS > 16);
391 masks = kmalloc(flen * sizeof(*masks), GFP_KERNEL);
394 memset(masks, 0xff, flen * sizeof(*masks));
396 for (pc = 0; pc < flen; pc++) {
397 memvalid &= masks[pc];
399 switch (filter[pc].code) {
402 memvalid |= (1 << filter[pc].k);
406 if (!(memvalid & (1 << filter[pc].k))) {
412 /* a jump must set masks on target */
413 masks[pc + 1 + filter[pc].k] &= memvalid;
416 case BPF_S_JMP_JEQ_K:
417 case BPF_S_JMP_JEQ_X:
418 case BPF_S_JMP_JGE_K:
419 case BPF_S_JMP_JGE_X:
420 case BPF_S_JMP_JGT_K:
421 case BPF_S_JMP_JGT_X:
422 case BPF_S_JMP_JSET_X:
423 case BPF_S_JMP_JSET_K:
424 /* a jump must set masks on targets */
425 masks[pc + 1 + filter[pc].jt] &= memvalid;
426 masks[pc + 1 + filter[pc].jf] &= memvalid;
437 * sk_chk_filter - verify socket filter code
438 * @filter: filter to verify
439 * @flen: length of filter
441 * Check the user's filter code. If we let some ugly
442 * filter code slip through kaboom! The filter must contain
443 * no references or jumps that are out of range, no illegal
444 * instructions, and must end with a RET instruction.
446 * All jumps are forward as they are not signed.
448 * Returns 0 if the rule set is legal or -EINVAL if not.
450 int sk_chk_filter(struct sock_filter *filter, unsigned int flen)
453 * Valid instructions are initialized to non-0.
454 * Invalid instructions are initialized to 0.
456 static const u8 codes[] = {
457 [BPF_ALU|BPF_ADD|BPF_K] = BPF_S_ALU_ADD_K,
458 [BPF_ALU|BPF_ADD|BPF_X] = BPF_S_ALU_ADD_X,
459 [BPF_ALU|BPF_SUB|BPF_K] = BPF_S_ALU_SUB_K,
460 [BPF_ALU|BPF_SUB|BPF_X] = BPF_S_ALU_SUB_X,
461 [BPF_ALU|BPF_MUL|BPF_K] = BPF_S_ALU_MUL_K,
462 [BPF_ALU|BPF_MUL|BPF_X] = BPF_S_ALU_MUL_X,
463 [BPF_ALU|BPF_DIV|BPF_X] = BPF_S_ALU_DIV_X,
464 [BPF_ALU|BPF_AND|BPF_K] = BPF_S_ALU_AND_K,
465 [BPF_ALU|BPF_AND|BPF_X] = BPF_S_ALU_AND_X,
466 [BPF_ALU|BPF_OR|BPF_K] = BPF_S_ALU_OR_K,
467 [BPF_ALU|BPF_OR|BPF_X] = BPF_S_ALU_OR_X,
468 [BPF_ALU|BPF_LSH|BPF_K] = BPF_S_ALU_LSH_K,
469 [BPF_ALU|BPF_LSH|BPF_X] = BPF_S_ALU_LSH_X,
470 [BPF_ALU|BPF_RSH|BPF_K] = BPF_S_ALU_RSH_K,
471 [BPF_ALU|BPF_RSH|BPF_X] = BPF_S_ALU_RSH_X,
472 [BPF_ALU|BPF_NEG] = BPF_S_ALU_NEG,
473 [BPF_LD|BPF_W|BPF_ABS] = BPF_S_LD_W_ABS,
474 [BPF_LD|BPF_H|BPF_ABS] = BPF_S_LD_H_ABS,
475 [BPF_LD|BPF_B|BPF_ABS] = BPF_S_LD_B_ABS,
476 [BPF_LD|BPF_W|BPF_LEN] = BPF_S_LD_W_LEN,
477 [BPF_LD|BPF_W|BPF_IND] = BPF_S_LD_W_IND,
478 [BPF_LD|BPF_H|BPF_IND] = BPF_S_LD_H_IND,
479 [BPF_LD|BPF_B|BPF_IND] = BPF_S_LD_B_IND,
480 [BPF_LD|BPF_IMM] = BPF_S_LD_IMM,
481 [BPF_LDX|BPF_W|BPF_LEN] = BPF_S_LDX_W_LEN,
482 [BPF_LDX|BPF_B|BPF_MSH] = BPF_S_LDX_B_MSH,
483 [BPF_LDX|BPF_IMM] = BPF_S_LDX_IMM,
484 [BPF_MISC|BPF_TAX] = BPF_S_MISC_TAX,
485 [BPF_MISC|BPF_TXA] = BPF_S_MISC_TXA,
486 [BPF_RET|BPF_K] = BPF_S_RET_K,
487 [BPF_RET|BPF_A] = BPF_S_RET_A,
488 [BPF_ALU|BPF_DIV|BPF_K] = BPF_S_ALU_DIV_K,
489 [BPF_LD|BPF_MEM] = BPF_S_LD_MEM,
490 [BPF_LDX|BPF_MEM] = BPF_S_LDX_MEM,
492 [BPF_STX] = BPF_S_STX,
493 [BPF_JMP|BPF_JA] = BPF_S_JMP_JA,
494 [BPF_JMP|BPF_JEQ|BPF_K] = BPF_S_JMP_JEQ_K,
495 [BPF_JMP|BPF_JEQ|BPF_X] = BPF_S_JMP_JEQ_X,
496 [BPF_JMP|BPF_JGE|BPF_K] = BPF_S_JMP_JGE_K,
497 [BPF_JMP|BPF_JGE|BPF_X] = BPF_S_JMP_JGE_X,
498 [BPF_JMP|BPF_JGT|BPF_K] = BPF_S_JMP_JGT_K,
499 [BPF_JMP|BPF_JGT|BPF_X] = BPF_S_JMP_JGT_X,
500 [BPF_JMP|BPF_JSET|BPF_K] = BPF_S_JMP_JSET_K,
501 [BPF_JMP|BPF_JSET|BPF_X] = BPF_S_JMP_JSET_X,
505 if (flen == 0 || flen > BPF_MAXINSNS)
508 /* check the filter code now */
509 for (pc = 0; pc < flen; pc++) {
510 struct sock_filter *ftest = &filter[pc];
511 u16 code = ftest->code;
513 if (code >= ARRAY_SIZE(codes))
518 /* Some instructions need special checks */
520 case BPF_S_ALU_DIV_K:
521 /* check for division by zero */
524 ftest->k = reciprocal_value(ftest->k);
530 /* check for invalid memory addresses */
531 if (ftest->k >= BPF_MEMWORDS)
536 * Note, the large ftest->k might cause loops.
537 * Compare this with conditional jumps below,
538 * where offsets are limited. --ANK (981016)
540 if (ftest->k >= (unsigned int)(flen-pc-1))
543 case BPF_S_JMP_JEQ_K:
544 case BPF_S_JMP_JEQ_X:
545 case BPF_S_JMP_JGE_K:
546 case BPF_S_JMP_JGE_X:
547 case BPF_S_JMP_JGT_K:
548 case BPF_S_JMP_JGT_X:
549 case BPF_S_JMP_JSET_X:
550 case BPF_S_JMP_JSET_K:
551 /* for conditionals both must be safe */
552 if (pc + ftest->jt + 1 >= flen ||
553 pc + ftest->jf + 1 >= flen)
559 #define ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \
560 code = BPF_S_ANC_##CODE; \
567 ANCILLARY(NLATTR_NEST);
573 ANCILLARY(ALU_XOR_X);
579 /* last instruction must be a RET code */
580 switch (filter[flen - 1].code) {
583 return check_load_and_stores(filter, flen);
587 EXPORT_SYMBOL(sk_chk_filter);
590 * sk_filter_release_rcu - Release a socket filter by rcu_head
591 * @rcu: rcu_head that contains the sk_filter to free
593 void sk_filter_release_rcu(struct rcu_head *rcu)
595 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
600 EXPORT_SYMBOL(sk_filter_release_rcu);
602 static int __sk_prepare_filter(struct sk_filter *fp)
606 fp->bpf_func = sk_run_filter;
608 err = sk_chk_filter(fp->insns, fp->len);
617 * sk_unattached_filter_create - create an unattached filter
618 * @fprog: the filter program
619 * @pfp: the unattached filter that is created
621 * Create a filter independent of any socket. We first run some
622 * sanity checks on it to make sure it does not explode on us later.
623 * If an error occurs or there is insufficient memory for the filter
624 * a negative errno code is returned. On success the return is zero.
626 int sk_unattached_filter_create(struct sk_filter **pfp,
627 struct sock_fprog *fprog)
629 struct sk_filter *fp;
630 unsigned int fsize = sizeof(struct sock_filter) * fprog->len;
633 /* Make sure new filter is there and in the right amounts. */
634 if (fprog->filter == NULL)
637 fp = kmalloc(fsize + sizeof(*fp), GFP_KERNEL);
640 memcpy(fp->insns, fprog->filter, fsize);
642 atomic_set(&fp->refcnt, 1);
643 fp->len = fprog->len;
645 err = __sk_prepare_filter(fp);
655 EXPORT_SYMBOL_GPL(sk_unattached_filter_create);
657 void sk_unattached_filter_destroy(struct sk_filter *fp)
659 sk_filter_release(fp);
661 EXPORT_SYMBOL_GPL(sk_unattached_filter_destroy);
664 * sk_attach_filter - attach a socket filter
665 * @fprog: the filter program
666 * @sk: the socket to use
668 * Attach the user's filter code. We first run some sanity checks on
669 * it to make sure it does not explode on us later. If an error
670 * occurs or there is insufficient memory for the filter a negative
671 * errno code is returned. On success the return is zero.
673 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
675 struct sk_filter *fp, *old_fp;
676 unsigned int fsize = sizeof(struct sock_filter) * fprog->len;
679 /* Make sure new filter is there and in the right amounts. */
680 if (fprog->filter == NULL)
683 fp = sock_kmalloc(sk, fsize+sizeof(*fp), GFP_KERNEL);
686 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
687 sock_kfree_s(sk, fp, fsize+sizeof(*fp));
691 atomic_set(&fp->refcnt, 1);
692 fp->len = fprog->len;
694 err = __sk_prepare_filter(fp);
696 sk_filter_uncharge(sk, fp);
700 old_fp = rcu_dereference_protected(sk->sk_filter,
701 sock_owned_by_user(sk));
702 rcu_assign_pointer(sk->sk_filter, fp);
705 sk_filter_uncharge(sk, old_fp);
708 EXPORT_SYMBOL_GPL(sk_attach_filter);
710 int sk_detach_filter(struct sock *sk)
713 struct sk_filter *filter;
715 filter = rcu_dereference_protected(sk->sk_filter,
716 sock_owned_by_user(sk));
718 RCU_INIT_POINTER(sk->sk_filter, NULL);
719 sk_filter_uncharge(sk, filter);
724 EXPORT_SYMBOL_GPL(sk_detach_filter);